Empirical Modeling of Bainite Formation Under Continuous Cooling Conditions Using Regression and Machine Learning Based Approaches

Bainitic and carbide‐free bainite advanced high‐strength steel grades have been of increasing interest, as they offer improved strength to ductility balances for high‐strength bar steel applications such as half shafts and crank shafts. To promote the development of grades that can form bainitic structures upon air cooling, this work presents a quantitative definition and empirical models of a steel's ability to form bainite, similar to how hardenability defines a steel's ability to form martensite. The critical cooling rates required for alloys to avoid martensite and ferrite/pearlite formation are recorded from 319 and 374 published continuous cooling transformation diagrams, respectively. Both a standard linear regression and a neural network model are created and validated from the data set, and the results of the two methods are compared. It is found that all alloying additions shift the bainitic shelf to the right, while only Mo and B additions are found to widen the bainitic shelf. The regression and neural network models predict similar behavior of the alloying elements tested, lending credibility to the use of machine learning for similar empirical modeling.